From d6aa1e2f65a4b8a6a3b9ea5fe43b33665835c30c Mon Sep 17 00:00:00 2001
From: Tyler Rockwood <rockwood@redpanda.com>
Date: Tue, 16 Apr 2024 01:42:03 +0000
Subject: [PATCH] wasm: fix CPU profiler hang

Currently trying to take a backtrace while the CPU profiler is running
results in a deadlock, because taking a backtrace segfaults, as we don't
write any debug symbols during JIT compilation (which causes other
deadlocks in libgcc). To fix this, disable the profile's backtracing
when within Wasm. In the future we should use Wasmtime's profiling APIs
to get a stacktrace within the guest program that is running. For more
information on that API see:
https://docs.wasmtime.dev/api/wasmtime/struct.GuestProfiler.html

Signed-off-by: Tyler Rockwood <rockwood@redpanda.com>
---
 src/v/wasm/tests/wasm_transform_test.cc | 19 ++++++++++++++++++-
 src/v/wasm/wasmtime.cc                  |  9 +++++++++
 2 files changed, 27 insertions(+), 1 deletion(-)

diff --git a/src/v/wasm/tests/wasm_transform_test.cc b/src/v/wasm/tests/wasm_transform_test.cc
index ae76ac359ea5a..d447e564d4ade 100644
--- a/src/v/wasm/tests/wasm_transform_test.cc
+++ b/src/v/wasm/tests/wasm_transform_test.cc
@@ -11,10 +11,10 @@
 
 #include "bytes/bytes.h"
 #include "pandaproxy/schema_registry/types.h"
-#include "serde/rw/rw.h"
 #include "wasm/errc.h"
 #include "wasm/tests/wasm_fixture.h"
 
+#include <seastar/core/internal/cpu_profiler.hh>
 #include <seastar/core/reactor.hh>
 
 #include <absl/strings/str_cat.h>
@@ -134,3 +134,20 @@ TEST_F(WasmTestFixture, LogsAreEmitted) {
     expected.append(reinterpret_cast<const char*>(bytes.data()), bytes.size());
     EXPECT_THAT(log_lines(), ElementsAre(expected));
 }
+
+TEST_F(WasmTestFixture, WorksWithCpuProfiler) {
+    bool original_enabled = ss::engine().get_cpu_profiler_enabled();
+    std::chrono::nanoseconds original_period
+      = ss::engine().get_cpu_profiler_period();
+    ss::engine().set_cpu_profiler_enabled(true);
+    ss::engine().set_cpu_profiler_period(10ns);
+    load_wasm("dynamic.wasm");
+    EXPECT_THROW(execute_command("loop", 0), wasm::wasm_exception);
+    ss::engine().set_cpu_profiler_enabled(original_enabled);
+    ss::engine().set_cpu_profiler_period(original_period);
+    std::vector<ss::cpu_profiler_trace> traces;
+    ss::engine().profiler_results(traces);
+    for (const auto& t : traces) {
+        std::cout << t.user_backtrace << "\n";
+    }
+}
diff --git a/src/v/wasm/wasmtime.cc b/src/v/wasm/wasmtime.cc
index e8fc7c70d8800..e7291252530dd 100644
--- a/src/v/wasm/wasmtime.cc
+++ b/src/v/wasm/wasmtime.cc
@@ -34,6 +34,7 @@
 
 #include <seastar/core/align.hh>
 #include <seastar/core/future.hh>
+#include <seastar/core/internal/cpu_profiler.hh>
 #include <seastar/core/posix.hh>
 #include <seastar/core/sharded.hh>
 #include <seastar/core/shared_ptr.hh>
@@ -599,7 +600,13 @@ class wasmtime_engine : public engine {
         // Poll the call future to completion, yielding to the scheduler when
         // the future yields.
         auto start = ss::steady_clock_type::now();
+        // Disable profiling backtraces inside the VM - at the time of writing
+        // backtraces lead to segfaults causing deadlock in Seastar's signal
+        // handlers.
+        auto _ = ss::internal::scoped_disable_profile_temporarily();
         while (!wasmtime_call_future_poll(fut.get())) {
+            // Re-enable stacktraces before we yield control to the scheduler.
+            ss::internal::profiler_drop_stacktraces(false);
             auto end = ss::steady_clock_type::now();
             _probe.increment_cpu_time(end - start);
             if (_pending_host_function) {
@@ -609,6 +616,8 @@ class wasmtime_engine : public engine {
                 co_await ss::coroutine::maybe_yield();
             }
             start = ss::steady_clock_type::now();
+            // Disable stacktraces as we enter back into Wasmtime
+            ss::internal::profiler_drop_stacktraces(true);
         }
         auto end = ss::steady_clock_type::now();
         _probe.increment_cpu_time(end - start);